World Development Vol. 93, pp. 75–93, 20170305-750X/� 2017 Elsevier Ltd. All rights reserved.

www.elsevier.com/locate/worlddevhttp://dx.doi.org/10.1016/j.worlddev.2016.12.020

Linking Economic Complexity, Institutions, and Income Inequality

DOMINIK HARTMANNa,b,c, MIGUEL R. GUEVARAa,d,e, CRISTIAN JARA-FIGUEROAa,MANUEL ARISTARAN a and CESAR A. HIDALGOa,*

aThe MIT Media Lab, Cambridge, USAbFraunhofer Center for International Management and Knowledge Economy IMW, Leipzig, Germany

cUniversity of Leipzig, GermanydUniversidad de Playa Ancha, Valparaıso, Chile

eUniversidad Tecnica Federico Santa Marıa, Valparaıso, Chile

Summary. — A country’s mix of products predicts its subsequent pattern of diversification and economic growth. But does this productmix also predict income inequality? Here we combine methods from econometrics, network science, and economic complexity to showthat countries exporting complex products—as measured by the Economic Complexity Index—have lower levels of income inequalitythan countries exporting simpler products. Using multivariate regression analysis, we show that economic complexity is a significantand negative predictor of income inequality and that this relationship is robust to controlling for aggregate measures of income, insti-tutions, export concentration, and human capital. Moreover, we introduce a measure that associates a product to a level of incomeinequality equal to the average GINI of the countries exporting that product (weighted by the share the product represents in that coun-try’s export basket). We use this measure together with the network of related products—or product space—to illustrate how the devel-opment of new products is associated with changes in income inequality. These findings show that economic complexity capturesinformation about an economy’s level of development that is relevant to the ways an economy generates and distributes its income.Moreover, these findings suggest that a country’s productive structure may limit its range of income inequality. Finally, we make ourresults available through an online resource that allows for its users to visualize the structural transformation of over 150 countriesand their associated changes in income inequality during 1963–2008.� 2017 Elsevier Ltd. All rights reserved.

Key words — income inequality, economic complexity, product space, institutions, economic development

1. INTRODUCTION

Is a country’s ability to both generate and distribute incomedetermined by its productive structure? Economic develop-ment pioneers, like Paul Rosenstein-Rodan, Hans Singer,and Albert Hirschman, would have said yes, since they arguedin favor of a connection between a country’s productive struc-ture, and its ability to generate and distribute income. Thesepioneers emphasized the economic role of ‘‘structural transfor-mations”—the process by which economies diversify fromagriculture and extractive industries to more sophisticatedforms of services and manufacturing (Hirschman, 1958;Rosenstein-Rodan, 1943; Singer, 1950).But testing the intuition of these development pioneers has

not been easy due to the complexity of measuring a country’sproductive structure. During the twentieth century, scholarsdid not go beyond simple quantitative approaches, such as(a) measuring the fraction of an economy employed in agricul-ture, manufacturing, or services; (b) using aggregate measuresof diversity and concentration (Herfindahl, 1950; Hirschman,1945; Imbs & Wacziarg, 2003); or (c) looking at diversificationinto related and unrelated varieties—that is, diversificationinto similar or different products (Boschma & Iammarino,2009; Frenken, Oort, & Verburg, 2007; Saviotti & Frenken,2008; Teece, Rumelt, Dosi, & Winter, 1994). These measuresof a country’s productive structure, however, fail to take thesophistication of the products into account, or capture differ-ences in industrial structures in a manner that is too coarse(i.e., by defining broad categories such as agriculture, manu-facturing, and services).Recently, though, the introduction of measures of ‘‘eco-

nomic complexity”—which we define and explain in the data

75

and methods section below—has expanded our ability toquantify a country’s productive structure and has revivedinterest in the macroeconomic role of structural transforma-tions (Abdon & Felipe, 2011; Bustos, Gomez, Hausmann, &Hidalgo, 2012; Caldarelli et al., 2012; Cristelli, Gabrielli,Tacchella, Caldarelli, & Pietronero, 2013; Cristelli,Tacchella, & Pietronero, 2015; Felipe, 2009; Hausmann,Hwang, & Rodrik, 2006; Hausmann et al., 2014; Hidalgo &Hausmann, 2009; Hidalgo, Klinger, Barabasi, & Hausmann,2007; Rodrik, 2006; Tacchella et al., 2012). These measuresof economic complexity have received wide attention becausethey are highly predictive of future economic growth. This alsomakes these measures of economic complexity relevant forsocial welfare, since economic growth and average incomeare correlated with country’s absolute levels of poverty andsocial welfare (Bourguignon, 2004; Ravallion, 2004).However, there are also multiple reasons why the productive

structures of countries could be associated not only with eco-nomic growth, but also with a country’s average level ofincome inequality.First, the mix of products that an economy makes con-

strains the occupational choices, learning opportunities, andbargaining power of its workers and unions. Notably, in sev-eral emerging economies, technological catch-up and industri-alization have provided new jobs and learning opportunitiesfor workers, contributing to the rise of a new middle class(Milanovic, 2012). Conversely in several ‘‘industrialized”economies, de-industrialization, de-unionization, and risingglobal competition for the export of industrial goods have

* Final revision accepted: December 17, 2016.

76 WORLD DEVELOPMENT

contributed to higher levels of income inequality. In the indus-trialized economies many industrial workers have becomeunemployed or were forced to work at low-paying jobs, andthe ability of unions to compress wage inequality hasdecreased (Acemoglu, Aghion, & Violante, 2001; Gustafsson& Johansson, 1999).Second, recent work on productive structures has high-

lighted that the complexity and diversity of products a countryexports are a good proxy of the knowledge and knowhowavailable in an economy that is not captured by aggregatemeasures of human capital (Hidalgo, 2015)—such as the yearsof schooling or the percentage of the population with tertiaryeducation. Moreover, productive structures can also be under-stood as a proxy of an economy’s level of social capital and thehealth of its institutions, since the ability of a country to pro-duce sophisticated products also critically depends on the abil-ity of people to form social and professional networks(Fukuyama, 1996; Hidalgo, 2015). For this reason, complexindustrial products also tend to require a large degree of tacitknowledge and more distributed knowledge than found withsimple products that are mainly based on resource richnessor low labor costs. More distributed knowledge and a largedegree of tacit knowledge can enhance the incentives to union-ize and increase the effectiveness in negotiating high wages andtherefore compress wage inequality.Third, in a world in which economic power begets political

power, non-diverse economies—such as countries withincomes largely based on few natural resources—are more sus-ceptible to suffer from both economic and political capture(Collier, 2007; Engerman & Sokoloff, 1997; Hartmann, 2014).Here, we contribute to the literature on economic complex-

ity, income inequality, and structural transformations, by doc-umenting a strong, robust, and stable correlation between acountry’s level of economic complexity (as proxied by the Eco-nomic Complexity Index) and its level of income inequalityduring 1963–2008. We find this correlation is robust to con-trolling for a variety of factors that are expected to explaincross-country variations in income inequality, such as a coun-try’s level of education, institutions, and export concentration.Moreover we find that, over time, countries that experienceincreases in economic complexity are more likely to experiencedecreases in their level of income inequality. We develop a pro-duct-level index to estimate the changes in the level of incomeinequality that we would expect if a country were to modify itsproduct mix by adding or removing a product. Our resultssuggest that a country’s level of income inequality may be con-ditioned by its productive structure.The remainder of the paper is structured as follows. Sec-

tion 2 reviews the literature on economic development, institu-tions, and income inequality. Section 3 presents the data andmethods used in this paper. Section 4 compares the correla-tions between Gini and a variety of measures of productivestructures, including the Economic Complexity Index(Hidalgo & Hausmann, 2009), the Herfindahl–HirschmanIndex (Herfindahl, 1950; Hirschman, 1945), Entropy(Shannon, 1948), and the Fitness Index (Tacchella et al.,2012). This section then uses multivariate regressions andpanel regressions to estimate the correlation between eco-nomic complexity and income inequality that is not explainedby the correlation between income inequality and averageincome, population, human capital (measured by averageyears of schooling), export concentration, and formal institu-tions. Finally, Section 5 introduces an estimator of the levelof income inequality expected for the exporters of 775 differentproducts in the Standard Industrial Trade Classification at thefour-digit level (SITC-4 Rev.2). We use this estimator to illus-

trate how changes in a country’s productive structure are asso-ciated with changes in income inequality. Section 6 providesconcluding remarks.

2. CONNECTING INCOME INEQUALITYAND ECONOMIC DEVELOPMENT

Decades ago Simon Kuznets (1955) proposed an inverted-u-shaped relationship describing the connection between a coun-try’s average level of income and its level of income inequality.Kuznets’ curve suggested that as an economy develops, marketforces would first increase and then decrease income inequal-ity. Yet, Kuznets’ curve has been difficult to verify. Theinverted-u-shaped relationship predicted by Kuznets fails tohold if several Latin American countries are removed fromthe sample (Deininger & Squire, 1998), and in recent decades,the upward side of Kuznets’ curve has vanished as inequalityin many low-income countries has increased (Palma, 2011).Moreover, several East-Asian economies have grown fromlow to middle incomes while reducing income inequality(Stiglitz, 1996). Together, these findings undermine the empir-ical robustness of Kuznets’ curve, and reaffirm that GDP percapita is an insufficient measure of economic development interms of explaining variations in income inequality (Kuznets,1934, 1973; Leontief, 1951; Stiglitz, Sen, & Fitoussi, 2009).The empirical failure of Kuznets’ curve resonates with

recent work arguing that inequality is not only dependent ona country’s rate or stage of growth, but also on its type ofgrowth and institutions (Acemoglu & Robinson, 2012;Beinhocker, 2006; Bourguignon, 2004; Collier, 2007;Engerman & Sokoloff, 1997; Fields; 2002; Hartmann, 2014;Ravallion, 2004; Sachs, 2005; Stiglitz et al., 2009). We shouldexpect, then, that more nuanced measures of economic devel-opment (such as those focused on the sophistication of theproducts that a country exports) should provide informationon the connection between economic development and incomeinequality that exceeds the limitations of aggregate outputmeasures like GDP.Understanding the determinants of income inequality is not

simple since income inequality depends on a variety of factors,from an economy’s factor endowments, geography, institu-tions, and social capital, to its historical trajectories, changesin technology, and returns to capital (Acemoglu et al., 2001;Acemoglu & Robinson, 2012; Autor, 2014; Beinhocker,2006; Brynjolfsson & McAfee, 2012; Collier, 2007; Davis,2009; Engerman & Sokoloff, 1997; Fields, 2002; Frey &Osborne, 2013; Gustafsson & Johansson, 1999; Hartmann,2014; Piketty, 2014; Stiglitz, 2013).Measuring these factors directly is difficult, but we can cre-

ate indirect measures of them by leveraging the fact that thepresence of these factors is expressed in a country’s mix ofproducts (Cristelli et al., 2013; Engerman & Sokoloff, 1997;Felipe, Kumar, Abdon, & Bacate, 2012; Hausmann &Rodrik, 2003; Hausmann et al., 2006, 2014; Hidalgo, 2015;Hidalgo & Hausmann, 2009; Hidalgo et al., 2007; Innis,1970; Rodrik, 2006; Tacchella et al., 2012;). For example,post-colonial economies specializing in a limited number ofagricultural or mineral products, like sugar, gold, and coffee,tend to have more unequal distributions of political power,human capital, and wealth (Acemoglu & Robinson, 2012;Engerman & Sokoloff, 1997; Innis, 1970), and hence, their pro-ductive structures provide us with indirect information abouttheir geographies, human capital, and institutions. Conversely,sophisticated products, like medical imaging devices or elec-tronic components, are typically produced in diversified

LINKING ECONOMIC COMPLEXITY, INSTITUTIONS, AND INCOME INEQUALITY 77

economies with inclusive institutions and high levels of humancapital. This means that the presence of complex industries inan economy, in addition to indicating the inclusiveness of thateconomy’s institutions, also reveals the knowledge and kno-whow that is embodied in its population (Hidalgo, 2015).The idea that productive structures co-evolve with the inclu-

siveness of institutions is not new, and can be traced back tothe work of scholars from the early twentieth century, likeHarold Innis (1970), and to more recent scholars, includingEngerman and Sokoloff (1997), or Acemoglu and Robinson(2012). Innis was a Canadian political economist who wroteextensively about how Canada’s early exports (mainly fur)helped determine Canada’s institutions (i.e., the Canadianrelationships with Native Americans and with Europe). Morerecently, Engerman and Sokoloff (1997) and Acemoglu andRobinson (2012) have built an institutional theory of interna-tional differences in income based on the idea that colonialpowers installed different institutions in their colonies.According to the theory, settlers installed extractive industriesand institutions when they found unfavorable conditions, butcreated the cities that homed both, non-extractive activitiesand inclusive institutions, when they found favorable condi-tions that allowed them to migrate in mass.From a modeling perspective, we can understand the co-

evolution between productive structures, institutions, andhuman capital, by assuming a model of heterogeneous firmsin which firms survive only when they are able to adopt or dis-cover the institutions and human capital that work best in theindustry that they participate in. This model assumes thatinstitutions are to a significant extent created at work anddepend on the type of industry. This assumption is extremelylikely, because on the one hand, people learn to interact andcollaborate with others in work settings, and on the other,there are clearly marked differences in the institutions (or cul-ture) of different sectors. For instance, the liberal institutionsthat are common in Silicon Valley’s tech sector might be idealfor industries that require workers to be creative problem sol-vers. These institutions, however, might be suboptimal in thecontext of a mining operation where following rules andrespecting hierarchies can ensure the safety of workers andthe coordination of the entire mining operation.Of course, there are often differences in the ways different

countries produce the same product. For instance, the sameindustry might be more labor intensive in one country andmore capital intensive in another country. However, thereare also significant differences in the particular skills, knowl-edge, factor endowments, and institutions that are needed tobecome globally competitive in a particular industry. Forinstance, the production of cocoa beans or coffee depends onthe availability of natural resources, while the production ofcomplex industries (like jet engines) depends on an extensivenetwork of skilled workers. We do not know a priori whetherdifferences among countries producing the same product (e.g.,the production of cars in Spain and Japan), are larger or smal-ler than differences among the production of different productsin the same country (the production of oranges and cars inSpain). Yet, the fact that we find a strong connection betweenproductive structures and income inequality suggests—butdoes not prove—that differences among the processes requiredto produce the same product in different countries, may besmaller than the differences among the processes required toproduce different products in the same countries.Therefore we argue in this paper that countries exporting

complex industries tend to be more inclusive and have lowerlevels of income inequality than countries that are exporting

simpler products. Some countries may be able to achieve com-parable high levels of average income based on naturalresources, but those comparably high levels of income willrarely come with inclusive institutions when they are not theresult of sophisticated industrial structures.A country’s productive structure can help explain variations

in institutions and income inequality, however, only if thereare significant differences in the productive structures ofmacro-economically similar countries. But how different arethe productive structures of macro-economically similar coun-tries? As an example consider Chile and Malaysia. In 2012,Chile’s average income per capita and years of schooling($21,044 at PPP in current 2012 US$ and 9.8 mean years ofschooling) was comparable to Malaysia’s income per capitaand schooling ($22,314 and 9.5). Yet, their productive struc-tures were practically orthogonal (see Figure 1), since Malay-sia’s exports mostly involved machinery and electronics, whileChile’s exports mostly involved agriculture and mining. TheEconomic Complexity Index (ECI) captures these differencesin productive structure. In 2012 Malaysia ranked 24th in theECI ranking while Chile ranked only 72nd (for the ECI rank-ings and further information about export structures seeatlas.media.mit.edu). Moreover, these differences in the ECIranking also point more accurately to differences in thesecountries’ level of income inequality. Chile’s inequality asmeasured through the Gini coefficient (GiniCHL = 0.49) is sig-nificantly higher than that of Malaysia (GiniMYS = 0.39),illustrating the correlation between income inequality and pro-ductive structures (see also Figure 2). The remainder of thepaper is dedicated to statistically testing this relationship fora large set of countries and years, as well as creating a pro-duct-level index of income inequality that allows for the visu-alization of the co-evolution between the structuraltransformation and income inequality for all countries inour dataset.

3. DATA AND METHODS

We use data from the Economic Complexity Index (ECI), aswell as international trade data, from MIT’s Observatory ofEconomic Complexity (atlas.media.mit.edu) (Simoes &Hidalgo, 2011). The trade data set combines exports data from1962 to 2000, compiled by Feenstra, Lipsey, Deng, Ma, andMo (2005), and data from the U.N. Comtrade for the periodduring 2001–12.The Economic Complexity Index (ECI) measures the

sophistication of a country’s productive structure by combin-ing information on the diversity of a country (the number ofproducts it exports), and the ubiquity of its products (the num-ber of countries that export that product) (Hidalgo &Hausmann, 2009). The intuition behind ECI is that sophisti-cated economies are diverse and export products that, on aver-age, have low ubiquity, because only a few diverse countriescan make these sophisticated products. By the same token, lesssophisticated economies are expected to produce a few ubiqui-tous products. ECI exploits this variation in the diversity ofcountries and the ubiquity of products to create a measureof a country’s productive structure that incorporates informa-tion about the sophistication of products.ECI is calculated from exports data connecting countries to

the products in which they have Revealed ComparativeAdvantages (RCA) (Hidalgo & Hausmann, 2009). TheRevealed Comparative Advantage (RCA) of a country c in aproduct p is:

Figure 1. Export structure of Chile (A) and Malaysia (B) in 2012. Source: atlas.media.mit.ed.

Figure 2. Bivariate relationships between economic complexity, income, and income inequality. Notes: All figures show that R2 and all p-values are less than

10–10. (A) ECI versus GINI EHII in 2000–08. (B) Natural logarithm of GDP per capita (constant 2005 US$) versus GINI EHII. (C) ECI versus GINI EHII

in 1963–69, (D) 1970–79, (E) 1980–89, and (F) 1990–99.

78 WORLD DEVELOPMENT

RCAcp ¼Xcp=

Pp0Xcp0P

c0Xc0p=P

c0p0Xc0p0

where Xcp is the total export of country c in product p. RCA islarger than 1 (indicating that a country has comparativeadvantage in a product), if a country’s export of a product

LINKING ECONOMIC COMPLEXITY, INSTITUTIONS, AND INCOME INEQUALITY 79

are larger than what would be expected from the size of thecountry’s export economy and the product’s global market.RCA is used to define a discrete matrix Mcp which is equal

to 1 if country c has RCA in product p and 0 otherwise.

Mcp ¼ 1 if RCAcp P 1

Mcp ¼ 0 if RCAcp < 1

The matrix Mcp allows to define the diversity of a countryand the ubiquity of a product, respectively, as the number ofproducts that are exported by a country with comparativeadvantage, and the number of countries that export a productwith comparative advantage.

Diversity ¼ kc0 ¼X

p

Mcp

Ubiquity ¼ kp0 ¼X

c

Mcp

Next, a matrix can be defined that connects countriesexporting similar products, weighted by the inverse of theubiquity of a product (to discount common products), andnormalized by the diversity of a country:

~Mcc0 ¼ 1

kc;0

X

p

McpMc0p

kp;0

Finally, the economic complexity index (ECI) is defined as

ECIc ¼ Kc � hKistdðKÞ

where Kc is the eigenvector of ~Mcc0 associated with the secondlargest eigenvalue—the vector associated with the largesteigenvalue is a vector of ones (Caldarelli et al., 2012;Hausmann et al., 2014; Kemp-Benedict, 2014).Table 1 shows the top five and bottom five economies based

on the ranking provided by ECI for the year 2012.Our income inequality data come from the Gini coefficient

estimates by Galbraith, Halbach, Malinowska, Shams, andZhang (2014) (GINI EHII dataset) and the GINI ALL datasetfrom Milanovic (2013). We mainly use the GINI EHII datasetbecause it is more comprehensive than other datasets (such asthe GINI ALL dataset from Milanovic (2013) for the periodduring 1963–89. The GINI EHII dataset contains inequalitydata for more than 60 countries starting in 1970, whereas

Table 1. Top and bottom five countries in t

ECI Rank Country

Top five countries by economic complexity1 Japan2 Switzerland3 Germany4 Sweden5 South Korea

ECI Rank ID

Bottom five countries by economic complexity120 Nigeria121 Turkmenistan122 Guinea123 Angola124 Libya

Source: atlas.media.mit.edu, World Bank’s World Development Indicators.

the GINI ALL dataset contains data for less than 40 countriesfor periods during 1963–89. Moreover the GINI ALL and isbiased toward countries with high levels of GDP per capitaand economic complexity (see Figure A1 in the Appendix).Additionally we use data on a country’s GDP per capita at

Purchasing Power Parity (PPP) in constant 2005 US$, averageyears of schooling, and population, from the World Bank’sWorld Development Indicators. Data on institutional vari-ables: corruption control, political stability, government effec-tiveness, regulatory quality, and voice and accountability,come from the World Bank’s Worldwide Governance Indica-tors (http://data.worldbank.org).It must be noted that we consider only countries with a pop-

ulation larger than 1.5 million and total exports of over 1 bil-lion dollars, thus removing small national economies that arecomparable to medium-size cities. The resulting datasetincludes 91% of the total world population and 84% of thetotal world trade during 1963–2008. Because of the sparsenessof several variables, especially the Gini data, we use averagevalues for the time periods 1963–69, 1970–79, 1980–89,1990–99, and 2000–08. We note that Gini values change rela-tively slow, so these averages are close to the Gini valuesexpected for each year within a decade. Summary statisticsfor all variables and each time period can be found in theAppendix (Table 4).

4. THE STATISTICAL CONNECTION BETWEEN ECO-NOMIC COMPLEXITY AND INCOME INEQUALITY

In this section we use bivariate and multivariate statistics toexplore the connection between economic complexity andincome inequality and to test its robustness to a number ofcontrols.

(a) Bivariate analysis: comparing the predictive power ofdifferent measures of productive structures

Figure 2 illustrates the bivariate relationship between Eco-nomic Complexity and Income Inequality in different decades,and compares the bivariate relationship between ECI-GINIwith the bivariate relationship between ECI and GDP per cap-ita (at Purchasing Power Parity, in constant 2005 US$) for 79countries during 2000–08. Both economic complexity andGDP per capita show a negative relationship with incomeinequality (Figure 2-A and -B). However, the negative

he Economic Complexity Ranking 2012

GDP per capita (cons 2005 US$) ECI

36,912 2.228858,557 1.973439,274 1.839645,260 1.708923,303 1.6999

GDP per capita (cons 2005 US$) ECI

1,034 �1.62893,270 �1.7245303 �1.78042,426 �2.24587,078 �3.1767

80 WORLD DEVELOPMENT

relationship between economic complexity and incomeinequality (R2 = 0.58, p-value = �10�16) is stronger than therelationship between income inequality and GDP per capita(R2 = 0.36, p-value = �10�10), and the difference in R2

between these two bivariate regressions is statistically signifi-cant. To test for the significance of the difference in R2 we usedthe Clarke-Test for non-nested models. Just as the F-test is thestandard method to select among nested models, the Clarke-test is the basic statistical model used to test the significanceof differences in R2 for non-nested models (models where theindependent variables are not perfect subsets of each other).In our case, the Clarke-Test prefers the ECI-GINI model overthe GDP per capita-GINI model with a p-value = 4.3 � 10�7

(also see Table 5 in the Appendix). Figure 2-C to -F, further-more, show that the negative bivariate relationship betweenincome inequality and economic complexity is stable acrossall considered decades. In all decades, we find a negative andsignificant relationship between economic complexity andincome inequality.Next we compare the bivariate relationships between differ-

ent measures of productive structures and income inequality inthe period 2000–08 (Figure 3). The matrix diagonal of Figure 3illustrates the histograms of each variable, the upper triangleof the matrix shows the correlation coefficients between eachpair of variables, and the lower triangle shows the correspond-ing scatterplots with a smoothed conditional mean line. ECIhas strong and significant correlations with all other measuresof productive structure, while ECI has a higher correlationwith the income inequality measures GINI EHII and GINIALL than GDP and all other measures of productive struc-tures. Table 5 and 6 in the Appendix also show the result ofthe Clarke tests comparing the predictive power of ECI forincome inequality with other measures of productive struc-tures for time periods. In the case of GINI EHII data, ECIis significantly preferred as predictor variable in 15 out of 16

Figure 3. Correlations between different economic diversity measures and

income inequality in 2000–08. HHI refers to the Herfindahl–Hirschman

Index and is a commonly used concentration measure; the Economic

Complexity Index (ECI), the Fitness Index, and Shannon Entropy are used

to measure the diversity and sophistication of a country’s exports.

model comparisons, whereas only in one model comparisonneither model is significantly preferred. In case of the GINIALL data, ECI is significantly preferred in 13 out of 16 modelcomparisons, while in three cases neither model is preferred.There is no case which income per capita or measures of pro-ductive structures are significantly preferred as predictor vari-able for income inequality in comparison to ECI.Next, we proceed to cross-sectional and panel regressions to

see if there is a significant correlation between economic com-plexity and inequality when controlling for other factors ofinequality like human capital or institutions. Afterward, wepresent a new measure that allows us to estimate the level ofinequality related to different types of products. This measure,in combination with the network of related products (Hidalgoet al., 2007), shows how the productive structure constrains acountry’s income inequality and opportunities for inclusiveeconomic development.

(b) Multivariate regression results

We start our analysis with a pooled regression for the periodfrom 1996 to 2008, and then we explore the changes in Ginis,during 1960s–2000s, using a panel regression for each decadethat includes country fixed effects. Because of the sparsenessof the Gini datasets and slow temporal changes in Ginis, weuse average values for each panel. We use the periods 1996–2001 and 2002–08 for cross-section regressions and 1963–69,1970–79, 1980–89, 1990–99, and 2000–08 for the fixed effectspanel regression. Due to the sparseness of the institutionalvariables, we only include them in the cross-section regres-sions.

(i) Pooled regressionTable 2 shows a pooled cross-sectional regression for the

periods of time between 1996–2001 and 2002–08. Columns1–6 illustrate a sequence of nested models that regress incomeinequality against economic complexity, GDP per capita atPurchasing Power Parity (PPP) and its square (a.k.a. Kuz-nets’ Curve), average years of schooling, population and theinstitutional factors: corruption control, government effective-ness, political stability, voice and accountability, and regulatoryquality.In every model, the Economic Complexity Index (ECI) is a

negative and significant predictor of income inequality. Edu-cation (as measured by average years of schooling) and logGDP squared also show a negative and significant correlationwith inequality; log GDP a positive and significant correlation.Notably, when we control for economic complexity, then therising part of the Kuznets curve is even more pronounced thanwithout it. Also the role of education in terms of years ofschooling becomes less important. Together, all variablesexplain 69.3% of the variance in income inequality amongcountries (Table 2, Column 1), but ECI is the most significantvariable in the regression analysis, and it is also the variablethat explains the largest percentage of the variance in incomeinequality after the effects of all other variables have beentaken into account. The semi-partial correlation of ECI (thedifference in R2 between the full model and one in which onlyECI was removed) is 8.1%, meaning that 8.1% of the variancein income inequality—which is not accounted for by institu-tional and macroeconomic variables—is explained by ECI(Table 2). Conversely the semi-partial correlations of all insti-tutional variables is less than 0.1%, while that of income, pop-ulation, and education, are all individually less than 2%. Thismeans that these variables capture information about inequal-ity that is already largely captured by ECI. Furthermore, ECI

Table 2. Pooled OLS regression models

Dependent variable: Gini

(I) (II) (III) (IV) (V) (VI)

ECI �0.040*** �0.037*** �0.046*** �0.033*** �0.044***

(0.007) (0.007) (0.007) (0.006) (0.006)ln(GDP PPP pc) 0.067** 0.059* 0.060** 0.056* 0.075***

(0.028) (0.032) (0.029) (0.028) (0.025)ln(GDP PPP pc)2 �0.004** �0.004* �0.003* �0.003* �0.004***

(0.002) (0.002) (0.002) (0.002) (0.001)Schooling �0.005*** �0.009*** �0.004** �0.006*** �0.005***

(0.002) (0.002) (0.002) (0.002) (0.002)Ln Population 0.007** 0.0001 0.005* 0.008*** 0.009***

(0.003) (0.003) (0.003) (0.003) (0.002)Rule of law �0.013 �0.016 �0.016 �0.015 �0.013

(0.013) (0.014) (0.013) (0.013) (0.013)Corruption Control 0.011 0.027* 0.009 0.016 0.007

(0.013) (0.014) (0.013) (0.013) (0.013)Government Effectiveness 0.002 �0.022 0.003 0.006 0.010

(0.017) (0.018) (0.017) (0.017) (0.017)Political stability �0.010 �0.017** �0.009 �0.009 �0.017***

(0.006) (0.007) (0.006) (0.006) (0.006)Regulatory quality �0.006 �0.012 �0.0002 �0.010 �0.012

(0.012) (0.014) (0.012) (0.012) (0.012)Voice and accountability 0.001 0.006 0.001 �0.004 0.003

(0.008) (0.009) (0.008) (0.008) (0.008)Constant 0.083 0.286** 0.391*** 0.068 0.244** 0.016

(0.130) (0.141) (0.050) (0.134) (0.114) (0.121)

Observations 142 142 142 142 142 142R2 0.717 0.639 0.701 0.699 0.704 0.704Adjusted R2 0.693 0.612 0.681 0.676 0.681 0.693Residual std. error 0.035 (df = 130) 0.039 (df = 131) 0.035 (df = 132) 0.035 (df = 131) 0.035 (df = 131) 0.035 (df = 136)F-statistic 29.916***

(df = 11; 130)23.208***

(df = 10; 131)34.413***

(df = 9; 132)30.458***

(df = 10; 131)31.165***

(df = 10; 131)64.656***

(df = 5; 136)

Notes: *p < 0.1; **p < 0.05; ***p < 0.01.These pooled OLS regression models regress income inequality against economic complexity, a country’s average level of income and its square,population, human capital, and the institutional variables: rule of law, corruption control, government effectiveness, political stability, regulatory quality,and voice and accountability. Column I includes all variables. Columns II–VI exclude blocks of variables to explore the contribution of each group ofvariables to the full model. The sharpest drop in R2 (from 0.693 to 0.612) is observed when ECI is removed from the regression. The table pools data fromtwo panels, one from 1996 to 2001 and another one from 2002 to 2008. The numbers in parenthesis are standard errors (SEM).

LINKING ECONOMIC COMPLEXITY, INSTITUTIONS, AND INCOME INEQUALITY 81

contains additional information about inequality that cannotbe explained by these other variables alone.In the Table 7–11 of the Appendix we also test these results

within each decade as well as using alternative Gini data sets(Galbraith et al., 2014; Milanovic, 2013) and alternative mea-sures of economic diversity, concentration, and fitness(Boschma & Iammarino, 2009; Cristelli et al., 2015; Frenkenet al., 2007; Herfindahl, 1950; Hirschman, 1945; Imbs &Wacziarg, 2003; Saviotti & Frenken, 2008; Tacchella et al.,2012; Teece et al., 1994). We find that our results are robustto these changes in datasets, methods, and classifications.

(ii) Temporal changesNext, we explore whether changes in a country’s level of

economic complexity are associated with changes in incomeinequality using a country-fixed-effect panel regression withdecade panels from 1963 to 2008. Unlike cross-sectionalresults, which make use of variations in inequality betweencountries, fixed-effect panel regressions exploit temporal vari-ations within a country. These variations are small for bothincome inequality and economic complexity, and thus weshould not expect large effects. Yet, despite the low levels oftemporal variation in the data, the fixed-effect panel regressionstill reveals a negative and significant association between a

country’s change in economic complexity and in its Gini coef-ficient (Table 3), meaning countries that experienced anincrease in economic complexity tended to experience adecrease in income inequality. In fact, we find that an increasein one standard deviation in economic complexity is associatedwith a reduction in Gini of 0.03. This association betweenchanges in economic complexity and income inequality isrobust to the inclusion of measures of income and human cap-ital. The institutional variables are not included since thosedata are only available for one of the five panels (the mostrecent one).This shows that the mix of products that a country exports

is a significant predictor of income inequality in both cross-sectional and panel regressions, even when controlling forother aggregated socioeconomic variables, like GDP, educa-tion, or population. Naturally, future research could addressvariations within educational achievements or variables likewithin-countries differences in access to infrastructure in moredetail (Acemoglu & Dell, 2010).While in this section we controlled the significant general

trends of the relationship between economic complexityand income inequality, the next section explores the inequal-ity related to 775 particular product categories and visualizesthe importance of the network structure of production—or

82 WORLD DEVELOPMENT

product space—for the subsequent diversification of coun-tries into more inclusive industries or less inclusive indus-tries.

5. THE PRODUCT SPACE AND INCOME INEQUALITY

This section first estimates the expected inequality related todifferent types of products and then uses this information tovisualize the country unique structural constraints on eco-nomic development and income inequality.

(a) Decomposing inequality at the product level

We decompose the relationship between economic complex-ity and income inequality into individual economic sectors bycreating a product-level estimator of the income inequalitythat is expected for the countries exporting a given product.We call this product level indicator the Product Gini Index(PGI). The PGI is closely related to previous measures onthe sophistication of exports—i.e., the export sophisticationmeasure of Lall, Weiss, and Zhang (2006), the PRODY ofHausmann et al. (2006) and the Product Complexity Indexof Hidalgo and Hausmann (2009)—showing that the type ofproducts a country exports determines its level of economicdevelopment (Hausmann et al., 2006; Hidalgo & Hausmann,2009; Hidalgo et al., 2007; Rodrik, 2006). However, insteadof relating products to income of the countries exporting theseproducts, the PGI explores the association of products withdifferent levels of income inequality.Decomposing income inequality at the product level can be

understood in the context of the co-evolution between produc-tive structures, education, and institutions, as discussed in theintroduction. To decompose income inequality at the productlevel we define the Product Gini Index (PGI) as the averagelevel of income inequality of a product’s exporters, weightedby the importance of each product in a country’s export bas-ket. Formally, we define the PGI (Product Gini Index) for aproduct p as:

Table 3. Fixed-effects

De

I II III

ECI �0.031*** �0.033*** �0.024****

�0.007 �0.007 �0.007ln(GDP PPP pc) �0.038 �0.042

�0.028 �0.027ln(GDP PPPpc)2 0.003* 0.002

�0.002 �0.002Schooling 0.010***

�0.002Ln population

Observations 338 338 338R2 0.077 0.123 0.198Adjusted R2 0.055 0.087 0.139F-statistic 20.13****

(df = 1; 240)11.14***

(df = 3; 238)14.63***

(df = 4; 237)

Country fixed effects Yes Yes Yes

Notes: *p < 0.1; **p < 0.05; ***p < 0.01.These seven fixed-effects panel regression models explore whether changes inincome inequality (column I), also controlling for the effects that other socioepopulation (column IV) have on income inequality. Columns V–VII control texcluded from the analysis. The numbers in parenthesis are standard errors (S

PGIp ¼ 1

Np

X

c

McpscpGinic

where Ginic is the Gini coefficient of country c,Mcp is 1 if countryc exports product p with revealed comparative advantage and 0otherwise, scp is the share of country c’s exports represented byproduct p. Np is a normalizing factor that ensures PGIs are theweighted average of the Ginis. Np and scp are calculated as:

Np ¼X

c

Mcpscp

where,

scp ¼ Xcp=X

p0Xcp0

where Xcp is the total export of product p by country c.We estimate PGIs using an average of Ginis for each pro-

duct, instead of using a regression with product dummies,because the number of products in our data is much largerthan the number of countries (e.g., 775 vs. 92 in 1995–2008),and hence, a regression would be over-specified.Figure 4A illustrates the construction of the PGI and Fig-

ure 4B shows the top 3, bottom 3 and median 3 productsaccording to the ranking of PGI values during 1995–2008(for all products see Table 15 in the Appendix). The productsassociated with the highest levels of income inequality (highPGI) mainly consist of commodities, such as Cocoa Beans,Inedible Flours of Meat and Fish, and Animal Hair. LowPGI products, on the other hand, include more sophisticatedforms of machinery and manufacturing products, such asPaper Making Machine Parts, Textile Machinery, and RoadRollers.Further information and descriptive statistics about the Pro-

duct Ginis can also be found in the Table 12–14 and Figure A2of the Appendix. It must be noted that products with a highlevel of economic complexity—measured by the product com-plexity index (Felipe et al., 2012; Hausmann et al., 2014;Hidalgo & Hausmann, 2009)—tend to also have lower PGI

panel regression

pendent variable: GINI

IV V VI VII

�0.026*** �0.030*** �0.029***

�0.007 �0.007 �0.007�0.017 �0.032 �0.053*

�0.029 �0.03 �0.03�0.00003 0.0005 0.004**

�0.002 �0.002 �0.0020.014*** 0.015*** 0.010***

�0.003 �0.003 �0.002�0.024** �0.016 �0.022** 0.014*

�0.011 �0.011 �0.01 �0.008

338 338 338 3380.213 0.165 0.196 0.1340.149 0.116 0.138 0.094

12.80***

(df = 5; 236)11.74***

(df = 4; 237)19.36***

(df = 3; 238)9.15***

(df = 4; 237)

Yes Yes Yes Yes

a country’s level of economic complexity are associated with changes inconomic factors like income (column II), human capital (column III) andhe variance explained by the model when ECI, income, or schooling, areEM).

LINKING ECONOMIC COMPLEXITY, INSTITUTIONS, AND INCOME INEQUALITY 83

values (Figure A2 of the Appendix). This means that complexproducts—such machine parts or electronic equipment forindustrial chains or I-Phones, robots, or 3D printingdevices—tend to be produced in more equalitarian countriesthan simpler and resource-exploiting products like cocoabeans or copper. It is common sense that complex productsrequire a larger network of skilled workers, related industries,and inclusive institutions making the economic competitive-ness of these products possible, than simpler industrial prod-ucts and resource exploiting activities whose competitivenessis mainly based on resource richness, low labor costs, rou-tinized activities, and economies of scale. A related observa-tion, which is important but beyond the scope of this paperis that these simpler products also tend to be located at thebeginning or the end of global production chains—they areeither extractive or assembly activities.Naturally, the association between product complexity and

income inequality also implies the need to understand the sys-temic distribution of inequality across the network of relatedproducts—or global product space (Hidalgo et al., 2007),and how productive transformations of the productive matrixare associated with changes in income inequality.

(b) The product space and the evolution of income inequality

In this section, we use PGIs in combination with the productspace—the network connecting products that are likely to beco-exported—to show how changes in a country’s productivestructure are connected to changes in a country’s level ofincome inequality.Figure 5A assigns colors for each product, using PGIs dur-

ing 1995–2008. Products associated with low levels of inequal-ity (low PGIs) are located in the center of the product space,where the more sophisticated products are located. On theother hand, high-PGI products tend to be located in theperiphery of the product space, where less sophisticated prod-ucts are located (6).We can also use the product space to study the constraints

on industrial diversification and the evolution of incomeinequality implied by a country’s productive structure. Theproduct space captures the notion that countries, cities, andregions, are significantly more likely to diversify toward prod-

Figure 4. The Product Gini Index (PGI). Notes: (A) The Product Gini Index (P

a product. The Gini coefficients of five copper exporters are shown in red. In blue

(B). Top three, middle three, and bottom three products by PGI values. The PGI

that contribute the most to each of these PGI are shown using diamonds. All va

references to color in this figure legend, the reader

ucts that are similar (i.e., connected in the product space) tothe products that they currently export (Boschma &Iammarino, 2009; Frenken et al., 2007; Hausmann et al.,2014; Hidalgo & Hausmann, 2009; Hidalgo et al., 2007;Neffke, Henning, & Boschma, 2011; Saviotti & Frenken,2008).Figure 5B–G compares the evolution of the productive

structure of Malaysia (5B–C), Norway (5D–E), and Chile(5F–G). Malaysia’s economy evolved from high-PGI productsin 1963–69—e.g., natural rubber, and sawlogs—to low-PGIproducts in 2000–08—e.g., electronic microcircuits and com-puter parts. Norway, on the other hand, moved in the oppositedirection, increasing its dependency on a high-PGI product—crude petroleum—and saw an increase in income inequality.Finally, Chile developed in a more constrained way, diversify-ing into products with a relatively high PGI—frozen fish, freshfish, and wine. More generally, these examples illustrate howthe productive structure of a country constrains the evolutionof its income inequality.

(c) Facilitating the interactive study of countries’ paths ofinclusive growth

Data on all countries can be explored in an interactive web-tool we designed for this paper. The web-tool is available atMIT’s observatory of economic complexity (atlas.media.mit.edu). The purpose of the interactive, visualization tool is toenable discussion among academicians, policy-makers, and prac-titioners about inclusive growth opportunities that takes intoaccount each country’s unique productive structure, particularopportunities for inclusive growth, and historical-structural-developments. The interactive online tool allows the user toexplore a map of 774 products between the years 1963 and 2014.

6. DISCUSSION/CONCLUDING REMARKS

Our results illustrate that the ability of an economy to bothgenerate and distribute income is strongly correlated with themix of products a country is able to produce and export.Taking economic complexity and productive space dynam-

ics into account allows us to reveal structural linkages between

GI) is a weighted average of the Gini coefficients of the countries that export

, we show the Gini coefficients of exporters of paper-making machine parts.

value is indicated with a black diamond. The Gini values of the five countries

lues are measured using data from 1995 to 2008. (For interpretation of the

is referred to the web version of this article.)

Figure 5. The product space and income inequality. (A) In this visualization of the product space nodes are colored according to a product’s PGI as measured

during 1995–2008. Node sizes are proportional to world trade during 2000–08. The networks are based on a proximity matrix representing 775 SITC-4 product

classes exported during 1963–2008. The link strength (proximity) is based on the conditional probability that the products are co-exported. (B) Malaysia’s

export portfolio during 1963–69. In this figure and the subsequent ones’ node sizes indicate the share of a product in a country’s export basket. Only products

with RCA greater than 1 are presented. (C) Malaysia’s export portfolio during 2000–08. (D) Norway’s exports during 1963–69 and (E) during 2000–08. (F)

Chile’s exports during 1963–69, and (G) during 2000–08.

84 WORLD DEVELOPMENT

economic development and income inequality which aggre-gated variables like the average years of schooling or incomeper capita alone are incapable of revealing. Our empiricalresults document a strong and robust correlation between

the economic complexity index and income inequality. Usingmultivariate regression, we confirmed that this relationship isrobust to controlling for measures of income, education, andinstitutions, and that this relationship has remained strong

Figure A1. Comparison between GINI EHII dataset (solid line) and GINI ALL dataset (dashed line). Notes: (A) Shows the total number of countries in

each dataset, by year. (B) Shows average ECI of countries in each dataset, by year.

LINKING ECONOMIC COMPLEXITY, INSTITUTIONS, AND INCOME INEQUALITY 85

over the last fifty years. Moreover, we showed that increases ineconomic complexity tend to be accompanied by decreases inincome inequality.Our findings do not mean that productive structures solely

determine a country’s level of income inequality. On the con-trary, a more likely explanation of the association between acountry’s productive structure and income inequality is that,as we argued in the introduction, productive structures repre-sent a high-resolution expression of a number of factors, frominstitutions to education, that co-evolve with a country’s mixof exported products and with the inclusiveness of its econ-omy. Still, because of this co-evolution, our findings empha-size the economic importance of productive structures, sincewe have shown that these are not only associated with incomeand economic growth (Hausmann et al., 2014; Hidalgo &Hausmann, 2009; Hidalgo et al., 2007), but also with howincome is distributed.Moreover, we advance methods that enable a more fine-

grained perspective on the relationship between productivestructures and income inequality. The method is based onintroducing the Product Gini Index or PGI, which estimatesthe expected level of inequality for the countries exporting agiven product. Overlaying PGI values on the network ofrelated products allows us to create maps that can be usedto anticipate how changes in a country’s productive structurecould affect its level of income inequality. These maps providemeans for researchers and policy-makers to explore and com-pare the co-evolution of productive structures, institutions,and income inequality for hundreds of economies.This paper and the related online tool show that a country’s

productive structure conditions its path of economic develop-ment and its abilities to generate and distribute income. Thisalso implies that social and industrial policies may need to com-plement each other to achieve sustained inequality reductionand economic development (Amsden, 2010; Hartmann, 2014;Ranis, Stewart, & Ramirez, 2000; Fajnzylber, 1990; Stiglitz,1996). While it is important for economic development andinequality reduction to improve school education and health ser-vices, it is also important to create advanced products and jobsthat demand specialized education and inclusive institutions.Of course, much more theoretical work needs to be done on

the complex relationships between economic complexity, insti-tutions, and income inequality. For example, there is a needfor more research on the importance of sector-wide institu-tions versus national institutions in different types of products.

Moreover, the role of unions and different types of knowl-edge—e.g., tacit or codified knowledge—in the relationshipbetween a country’s productive matrix and its level of incomeinequality needs to be more thoroughly explored. Finally, theassociation between industrialization and gender equalityrequires further research.More in-depth case studies are also required, for example in

countries like Mexico and Australia. Mexico shows both a highlevel of income inequality and a high ECI. This might be due tothe impact of US production facilities and a slightly inflated ECIvalue, a fact that is partly expressed in the lack of diversity ofMexico’s export destinations. Conversely, Australia has partlysucceeded in establishing inclusive institutions with relativelylow levels of income inequality, despite having an export struc-ture focused on natural resource extraction. Yet, there may beother activities that are not captured in trade data—such asthe export of mining services and operations—that may helppartly explain the anomaly. In other words, Australia mightbe a case where the measure of complexity based on exportsunderestimates the country’s real complexity.We must also note that our results are limited to the time

period during 1963–2008. Yet, significant root causes of thepresent institutions, world production system, and distributionof income can be traced back to the Commercial Revolutionand colonization in the 15th–18th century, as well as theIndustrial Revolution that has started in the 18th centuryand has subsequently spread across the world. Certainly notall segments of the population have always benefitted froman increase in economic complexity, since many individualssuffered—or continue to suffer—from unemployment, slavery,or exploitative working conditions. Nonetheless, the socialinstitutions necessary to allow countries to use new technolo-gies, sustain creative destruction processes, and achieve highlevels of economic complexity over the middle to long-run,have tended to include larger segments of the population backagain into the economic development processes, at least in theindustrially most advanced economies (Acemoglu &Robinson, 2012; Perez, 2003).Despite the limitations, we showed that applying methods

from economic complexity and network science allows us tocapture significant information about countries’ path of eco-nomic development that goes beyond aggregated input or out-put factors like GDP or years of schooling. The developmentpioneers highlighted the role of productive structures andstructural transformations in economic development. Here

86 WORLD DEVELOPMENT

we have introduced methods that are capable of exploring howheterogeneous productive structures of countries conditiontheir ability to generate and distribute income in more detail.Moreover, we can use these methods to further explore the

association between economic complexity and the evolutionof institutions.In sum, our analysis strongly suggests that countries exporting

more complex products tend to have significantly lower levels ofincome inequality than countries exporting simple products.

REFERENCES

Abdon, A., & Felipe, J. (2011). The product space. What does it say aboutthe opportunities for growth and structural transformation of sub-Saharan Africa? (Economics Working Paper Archive No. wp_670).Levy Economics Institute, Retrieved from https://ideas.repec.org/p/lev/wrkpap/wp_670.html.

Acemoglu, D., Aghion, P., & Violante, G. L. (2001). Deunionization,technical change and inequality (SSRN Scholarly Paper No. ID 267264). Rochester, NY: Social Science Research Network, Retrieved fromhttp://papers.ssrn.com/abstract=267264.

Acemoglu, D., & Dell, M. (2010). Productivity differences between andwithin countries. American Economic Journal: Macroeconomics, 2(1),169–188.

Acemoglu, D., & Robinson, J. (2012). Why nations fail: The origins ofpower, prosperity, and poverty. New York: Crown.

Amsden, A. H. (2010). Say’s law, poverty persistence, and employmentneglect. Journal of Human Development and Capabilities, 11(1), 57–66.

Autor, D. H. (2014). Skills, education, and the rise of earnings inequalityamong the ‘‘other 99 percent”. Science, 344(6186), 843–851. http://dx.doi.org/10.1126/science.1251868.

Beinhocker, E. D. (2006). The origin of wealth: Evolution, complexity, andthe radical remaking of economics. Harvard Business Press.

Boschma, R., & Iammarino, S. (2009). Related variety, trade linkages, andregional growth in Italy. Economic Geography, 85(3), 289–311. http://dx.doi.org/10.1111/j.1944-8287.2009.01034.x.

Bourguignon, F. (2004). The poverty-growth-inequality triangle (IndianCouncil for Research on International Economic Relations, New DelhiWorking Paper No. 125). New Delhi: Indian Council for Research onInternational Economic Relations, Retrieved from https://ideas.repec.org/p/ind/icrier/125.html.

Brynjolfsson, E., & McAfee, A. (2012). Race against the machine: How thedigital revolution is accelerating innovation, driving productivity, andirreversibly transforming employment and the economy. Lexington, MA:Digital Frontier Press.

Bustos, S., Gomez, C., Hausmann, R., & Hidalgo, C. A. (2012). Thedynamics of nestedness predicts the evolution of industrial ecosystems.PLoS One, 7(11), e49393. http://dx.doi.org/10.1371/journal.pone.0049393.

Caldarelli, G., Cristelli, M., Gabrielli, A., Pietronero, L., Scala, A., &Tacchella, A. (2012). A network analysis of countries’ export flows:Firm grounds for the building blocks of the economy. PLoS One, 7(10), e47278. http://dx.doi.org/10.1371/journal.pone.0047278.

Collier, P. (2007). The bottom billion: Why the poorest countries are failingand what can be done about it. New York: Oxford Univ. Press.

Cristelli, M., Gabrielli, A., Tacchella, A., Caldarelli, G., & Pietronero, L.(2013). Measuring the intangibles: A metrics for the economiccomplexity of countries and products. PLoS One, 8(8), e70726.http://dx.doi.org/10.1371/journal.pone.0070726.

Cristelli, M., Tacchella, A., & Pietronero, L. (2015). The heterogeneousdynamics of economic complexity. PLoS One, 10(2), e0117174. http://dx.doi.org/10.1371/journal.pone.0117174.

Davis, G. F. (2009). Managed by the markets: How finance re-shapedAmerica. Oxford: Oxford Univ. Press.

Deininger, K., & Squire, L. (1998). New ways of looking at old issues:Inequality and growth. Journal of Development Economics, 57(2),259–287. http://dx.doi.org/10.1016/S0304-3878(98)00099-6.

Engerman, S. L., & Sokoloff, K. L. (1997). Factor endowments,institutions, and differential paths of growth among new worldeconomies. In S. H. Haber (Ed.), How Latin America fell behind:Essays on the economic histories of Brazil and Mexico, 1800–1914(pp. 260–304). California: Stanford Univ. Press.

Fajnzylber, F. (1990). Industrialization in Latin America: From the ‘BlackBox’ to the ‘Empty Box’: A comparison of contemporary industrializa-tion patterns, Cuadernos de la CEPAL, No. 60, 1990-08. , Retrievedfrom http://hdl.handle.net/11362/27811.

Feenstra, R. C., Lipsey, R. E., Deng, H., Ma, A. C., & Mo, H. (2005).World trade flows: 1962-2000 (Working Paper No. 11040). NationalBureau of Economic Research, Retrieved from http://www.nber.org/papers/w11040.

Felipe, J. (2009). Inclusive growth, full employment and structural change:Implications and policies for developing Asia. London: Anthem Press.

Felipe, J., Kumar, U., Abdon, A., & Bacate, M. (2012). Productcomplexity and economic development. Structural Change and Eco-nomic Dynamics, 23(1), 36–68. http://dx.doi.org/10.1016/j.strue-co.2011.08.003.

Frenken, K., Oort, F. V., & Verburg, T. (2007). Related variety, unrelatedvariety and regional economic growth. Regional Studies, 41(5),685–697. http://dx.doi.org/10.1080/00343400601120296.

Frey, C. B., & Osborne, M. A. (2013). The future of employment: Howsusceptible are jobs to computerisation?. Oxford Martin School,Retrieved from: http://www.oxfordmartin.ox.ac.uk/publications/view/1314.

Fukuyama, F. (1996). Trust: Human nature and the reconstitution of socialorder. New York: Free Press.

Galbraith, J. K., Halbach, B., Malinowska, A., Shams, A., & Zhang, W.(2014). UTIP global inequality data sets 1963-2008: Updates, revisionsand quality checks (Working Paper No. 68). University of Texas atAustin, Retrieved from http://utip.gov.utexas.edu/papers/UTIP_68.pdf.

Gustafsson, B., & Johansson, M. (1999). In search of smoking guns: Whatmakes income inequality vary over time in different countries?.American Sociological Review, 64(4), 585–605. http://dx.doi.org/10.2307/2657258.

Hartmann, D. (2014). Economic complexity and human development: Howeconomic diversification and social networks affect human agency andwelfare. New York: Routledge.

Hausmann, R., Hidalgo, C. A., Bustos, S., Coscia, M., Simoes, A., &Yildirim, M. A. (2014). The atlas of economic complexity: Mappingpaths to prosperity. MIT Press.

Hausmann, R., Hwang, J., & Rodrik, D. (2006). What you export matters.Journal of Economic Growth, 12(1), 1–25. http://dx.doi.org/10.1007/s10887-006-9009-4.

Hausmann, R., & Rodrik, D. (2003). Economic development as self-discovery. Journal of Development Economics, 72(2), 603–633. http://dx.doi.org/10.1016/S0304-3878(03)00124-X.

Herfindahl, O. C. (1950). Concentration in the steel industry. .Hidalgo, C. (2015). Why information grows: The evolution of order, from

atoms to economies. New York: Penguin Press.Hidalgo, C. A., & Hausmann, R. (2009). The building blocks of economic

complexity. Proceedings of the National Academy of Sciences, 106(26),10570–10575. http://dx.doi.org/10.1073/pnas.0900943106.

Hidalgo, C. A., Klinger, B., Barabasi, A.-L., & Hausmann, R. (2007). Theproduct space conditions the development of nations. Science, 317(5837), 482–487. http://dx.doi.org/10.1126/science.1144581.

Hirschman, A. O. (1945). National power and the structure of foreign trade.Berkeley and Los Angeles: University of California Press.

Hirschman, A. O. (1958). The strategy of economic development (Vol. 10)New Haven: Yale Univ. Press.

Imbs, J., & Wacziarg, R. (2003). Stages of diversification. The AmericanEconomic Review, 93(1), 63–86.

Innis, H. A. (1970). The fur trade in Canada: An introduction to Canadianeconomic history. Univ. of Toronto Press.

Kemp-Benedict, E. (2014). An interpretation and critique of the method ofreflections (No. 60705). , Retrieved from https://www.sei-international.org/publications?pid=2636.

Kuznets, S. (1934). National income, 1929-1932. In National income, 1929-1932. NBER.

Kuznets, S. (1955). Economic growth and income inequality. TheAmerican Economic Review, 45(1), 1–28.

Kuznets, S. (1973). Modern economic growth: Findings and reflections.The American Economic Review, 63(3), 247–258.

LINKING ECONOMIC COMPLEXITY, INSTITUTIONS, AND INCOME INEQUALITY 87

Lall, S., Weiss, J., & Zhang, J. (2006). The ‘‘sophistication” of exports: Anew trade measure. World Development, 34(2), 222–237. http://dx.doi.org/10.1016/j.worlddev.2005.09.002.

Leontief, W. W. (1951). Input–output economics. Scientific American, 185(4), 15–21.

Milanovic, B. (2012). Global income inequality by the numbers: In historyand now –An overview. The World Bank, Retrieved from http://elibrary.worldbank.org/doi/abs/10.1596/1813-9450-6259.

Milanovic, B. L. (2013). All the Ginis dataset. , Retrieved June 28, 2015,from http://data.worldbank.org/data-catalog/all-the-ginis.

MIT’s Observatory of Economic Complexity: Country Ranking. (n.d.).Retrieved June 28, 2015, from https://atlas.media.mit.edu/rankings.

Neffke, F., Henning, M., & Boschma, R. (2011). How do regions diversifyover time? Industry relatedness and the development of new growthpaths in regions. Economic Geography, 87(3), 237–265. http://dx.doi.org/10.1111/j.1944-8287.2011.01121.x.

Palma, J. G. (2011). Homogeneous middles vs. heterogeneous tails, andthe end of the ‘‘Inverted-U”: It’s all about the share of the rich.Development and Change, 42(1), 87–153. http://dx.doi.org/10.1111/j.1467-7660.2011.01694.x.

Perez, C. (2003). Technological revolutions and financial capital: Thedynamics of bubbles and golden ages. Cheltenham, UK; Northampton,MA, USA: Edward Elgar Pub..

Piketty, T. (2014). Capital in the twenty-first century (A. Goldhammer,Trans.). Cambridge, MA: Belknap Press.

Ranis, G., Stewart, F., & Ramirez, A. (2000). Economic growth andhuman development’. World Development, 28(2), 197–219.

Ravallion, M. (2004). Pro-poor growth: A primer (Policy ResearchWorking Paper No. ID 610283). World Bank, Retrieved from http://dx.doi.org/10.1596/1813-9450-3242.

Rodrik, D. (2006). What’s so special about China’s exports?. China &World Economy, 14(5), 1–19. http://dx.doi.org/10.1111/j.1749-124X.2006.00038.x.

Rosenstein-Rodan, P. N. (1943). Problems of industrialisation of easternand south-eastern Europe. The Economic Journal, 53(210/211),202–211. http://dx.doi.org/10.2307/2226317.

Sachs, J. (2005). The end of poverty: How we can make it happen in ourlifetime. New York: Penguin Books Limited.

Saviotti, P. P., & Frenken, K. (2008). Export variety and the economicperformance of countries. Journal of Evolutionary Economics, 18(2),201–218. http://dx.doi.org/10.1007/s00191-007-0081-5.

Shannon, C. E. (1948). A mathematical theory of communication. TheBell System Technical Journal, 27, 379-423-656.

Simoes, A. J. G., & Hidalgo, C. A. (2011). The Economic ComplexityObservatory: An analytical tool for understanding the dynamics ofeconomic development. Workshops at the Twenty-Fifth AAAI Con-ference on Artificial Intelligence 2011.

Singer, H. W. (1950). The distribution of gains between investing andborrowing countries. The American Economic Review, 40(2), 473–485.

Stiglitz, J. E. (1996). Some lessons from the East Asian miracle. The WorldBank Research Observer, 11(2), 151–177. http://dx.doi.org/10.1093/wbro/11.2.151.

Stiglitz, J. E. (2013). The price of inequality: How today’s divided societyendangers our future (1st ed.) New York: W. W. Norton.

Stiglitz, J., Sen, A. K., & Fitoussi, J.-P. (2009). The measurement ofeconomic performance and social progress revisited: reflections andoverview. , Retrieved from https://hal-sciencespo.archives-ouvertes.fr/hal-01069384/document.

Teece, D., Rumelt, R., Dosi, G., & Winter, S. (1994). Understandingcorporate coherence: Theory and evidence. Journal of EconomicBehavior & Organization, 23(1), 1–30.

APPENDIX

Descriptive statistics

Robustness check for different inequality measures

We find that ECI continues to be a negative and significantpredictor of income inequality when the GINI ALL dataset isused instead of GINI EHII dataset (see Table 7). However, logGDP gains in significance as an inequality predictor. As was

previously discussed, the GINI ALL dataset is strongly biasedtoward countries with a complex economy (see Figure A1),therefore is not a surprise that ECI loses some of its predictivepower, since there is not much variation in ECI between coun-tries in the sample.Next, we control the robustness of cross-sectional results

within each decade. Table 8 shows that there is a negativeand significant relationship between economic complexityand GINI EHII in all decades. Moreover, Table 9 presentsthe results of the same regression using GINI ALL as thedependent variable. In all decades, and in both tables, ECIresults to be a negative and significant predictor on incomeinequality. The importance of GDP increases when using theGINI ALL dataset, but ECI continues to explain a significantpercentage of the variance of income inequality.

Effects of other measures of export diversity, complexity, andconcentration

Table 10 reproduces the cross-sectional regression Table 1from the paper, with the addition of the Fitness Index, Shan-non Entropy, and the Herfindahl–Hirschman Index (HHI).The results show that all of these measures are significantwhen included in the regression individually, however ECIexplains a larger fraction of the variance in inequality. Nota-bly, a higher economic concentration seems to lead to a higherlevel of income inequality, however this effect is not significantif economic complexity (ECI) is included.Next, we compare the effects of ECI (Hidalgo & Hausmann,

2009), the Fitness Index (Tacchella et al., 2012), Shannon-Entropy (Shannon, 1948), and the Herfindahl–HirschmanIndex (Herfindahl, 1950; Hirschman, 1945) in fixed-effectspanel regression (Table 11). The results show that, amongthese measures of productive structure, ECI is the only mea-sure which is a significant predictor of time variations ininequality within countries over long periods of time.

Product Gini index

This section analyzes the evolution of PGIs over time, andto which extent complex products are also more inclusiveproducts.

PGI rankingTable 12 shows the highest and lowest ranked five products

in the PGI index during 1995–2008. A full list of all 775 prod-ucts can be found in a separate Appendix Table 15.

Descriptive statistics and evolution of PGIsAs discussed in the paper, the PGIs associate each product

with a level of income inequality by calculating the averageGini coefficient of the countries that produce the respectiveproduct, weighted by the product’s importance in the coun-try’s economy. Table 13 shows summary statistics for the PGIsfor each decade intervals. Due to limited data availability weexclude the period during 1963–69.The average value of the PGIs increases over time, repre-

senting the trend found in recent research on inequality mea-sures that countries are converging toward an average Ginivalue around 0.40 (Palma, 2011). However, despite this con-vergence trend, the spread between minimum and maximumvalue of the PGIs remains large. While the PGI value variedbetween 0.285 and 0.511 in the time period during 1970–79,in the time period during 2000–08 the values were distributedbetween 0.334 and 0.517.

Table 4. Descriptive summary statistics

Variable Year Obs. Mean SD Min Max

ECI 1963–69 102 0.01 1.11 �2.67 2.16ECI 1970–79 108 �0.03 1.03 �2.51 1.90ECI 1980–89 103 0.00 1.07 �2.17 2.18ECI 1990–99 125 0.02 1.02 �1.79 2.31ECI 2000–08 128 0.00 0.98 �1.75 2.54ECI All decades 566 0.00 1.03 �2.67 2.54GINI EHII 1963–69 85 42.67 7.35 22.32 54.59GINI EHII 1970–79 108 41.65 7.66 21.22 52.49GINI EHII 1980–89 121 41.48 7.59 20.93 53.09GINI EHII 1990–99 125 43.49 6.94 28.10 58.25GINI EHII 2000–08 103 43.69 6.40 29.62 55.02GINI EHII All decades 542 42.59 7.24 20.93 58.25GINI All 1963–69 44 41.46 10.06 20.88 62.00GINI All 1970–79 62 40.30 9.05 21.80 61.25GINI All 1980–89 99 36.31 11.24 18.60 62.90GINI All 1990–99 136 40.52 10.23 20.49 74.30GINI All 2000–08 148 40.05 9.55 23.98 72.95GINI All All decades 489 39.58 10.19 18.60 74.30GDP PPP05 pc 1963–69 71 4,706.9 5,984.5 113.0 2,1871.4GDP PPP05 pc 1970–79 76 6,423.5 7,881.9 168.8 2,9260.0GDP PPP05 pc 1980–89 92 8,236.6 11,578.5 150.1 59,411.1GDP PPP05 pc 1990–99 115 8,603.9 12,061.4 128.2 51,800.0GDP PPP05 pc 2000–08 123 11,030.6 14,669.9 157.6 64,311.1GDP PPP05 pc All decades 477 8,231.4 11,624.0 113.0 64,311.1ln(GDP PPP05 pc) 1963–69 71 7.64 1.34 4.73 9.99ln(GDP PPP05 pc) 1970–79 76 7.93 1.39 5.13 10.28ln(GDP PPP05 pc) 1980–89 92 7.97 1.56 5.01 10.99ln(GDP PPP05 pc) 1990–99 115 8.04 1.52 4.85 10.86ln(GDP PPP05 pc) 2000–08 123 8.33 1.51 5.06 11.07ln(GDP PPP05 pc) All decades 477 8.02 1.49 4.73 11.07ln(GDP PPP05 pc)2 1963–69 71 60.08 20.83 22.35 99.86ln(GDP PPP05 pc)2 1970–79 76 64.76 22.20 26.30 105.76ln(GDP PPP05 pc)2 1980–89 92 65.94 25.15 25.11 120.83ln(GDP PPP05 pc)2 1990–99 115 66.86 24.88 23.56 117.83ln(GDP PPP05 pc)2 2000–08 123 71.68 25.35 25.60 122.58ln(GDP PPP05 pc)2 All decades 477 66.58 24.27 22.35 122.58Years of schooling 1963–69 81 3.63 2.56 0.44 10.17Years of schooling 1970–79 85 4.16 2.72 0.64 11.16Years of schooling 1980–89 86 4.85 2.82 0.85 11.80Years of schooling 1990–99 102 6.35 2.86 0.66 12.26Years of schooling 2000–08 105 7.32 2.84 0.88 12.92Years of schooling All decades 459 5.41 3.09 0.44 12.92Population (million) 1963–69 93 67.0 361.9 1.1 3,401.4Population (million) 1970–79 96 77.9 422.6 1.0 4,027.0Population (million) 1980–89 101 87.2 468.1 1.0 4,560.0Population (million) 1990–99 119 44.8 142.0 1.1 1,196.0Population (million) 2000–08 125 48.2 155.1 1.0 1294.4Corruption control 1990–99 110 0.08 1.06 �1.28 2.40Corruption control 2000–08 116 0.08 1.03 �1.48 2.48Corruption control All decades 226 0.08 1.04 �1.48 2.48Government effectiveness 1990–99 110 0.13 0.97 �1.25 2.12Government effectiveness 2000–08 116 0.15 0.97 �1.51 2.18Government effectiveness All decades 226 0.14 0.97 �1.51 2.18Political stability 1990–99 110 �0.05 0.90 �2.38 1.51Political stability 2000–08 116 �0.06 0.89 �2.08 1.58Political stability All decades 226 �0.05 0.89 �2.38 1.58Regulatory quality 1990–99 110 0.14 0.95 �2.04 2.21Regulatory quality 2000–08 116 0.15 0.95 �2.16 1.89Regulatory quality All decades 226 0.15 0.95 �2.16 2.21Rule of law 1990–99 110 0.01 1.00 �1.67 1.93Rule of law 2000–08 116 0.03 0.98 �1.53 1.93Rule of law All decades 226 0.02 0.99 �1.67 1.93Voice and accountability 1990–99 110 0.02 0.97 �1.95 1.67Voice and accountability 2000–08 116 0.00 1.00 �2.13 1.63

(continued on next page)

88 WORLD DEVELOPMENT

Table 5. Clarke test for Gini EHII

Model 1 Model 2 Year Clarke Test-Stat

ECI Fitness Index Pooled Model 1 is preferred (p < 2e�16) 302 (81%)ECI Fitness Index 2000–08 Model 1 is preferred (p = 9.1e�05) 62 (71%)ECI Fitness Index 1990–99 Model 1 is preferred (p = 2.5e�05) 67 (72%)ECI Fitness Index 1980–89 Model 1 is preferred (p = 1.0e�07) 59 (81%)ECI Entropy Pooled Model 1 is preferred (p < 2e�16) 287 (77%)ECI Entropy 2000–08 Model 1 is preferred (p = 4.3e�06) 65 (75%)ECI Entropy 1990–99 Model 1 is preferred (p = 2.5e�05) 67 (72%)ECI Entropy 1980–89 Model 1 is preferred (p = 0.00037) 52 (71%)ECI HHI Pooled Model 1 is preferred (p < 2e�16) 296 (80%)ECI HHI 2000–08 Model 1 is preferred (p = 4.3e�06) 65 (75%)ECI HHI 1990–99 Model 1 is preferred (p = 4.4e�10) 76 (82%)ECI HHI 1980–89 Model 1 is preferred (p = 4.1e�07) 58 (79%)ECI Log GDP Pooled Model 1 is preferred (p = 2.8e�15) 261 (70%)ECI Log GDP 2000–08 Model 1 is preferred (p = 4.3e�07) 67 (77%)ECI Log GDP 1990–99 Model 1 is preferred (p = 3.5e�07) 71 (76%)ECI Log GDP 1980–89 Neither model is significantly preferred (p = 0.82) 38 (52%)

Notes: Clarke test compares Model 1 (ECI) with different measures of productive structure (Model 2). The dependent variable is GINI EHII.

Table 6. Clarke test for Gini All

Model 1 Model 2 Year Clarke Test-Stat

ECI Fitness Index Pooled Model 1 is preferred (p = 8.4e�14) 248 (70%)ECI Fitness Index 2000–08 Neither model is significantly preferred (p = 1) 52 (50%)ECI Fitness Index 1990–99 Model 1 is preferred (p = 0.0039) 66 (65%)ECI Fitness Index 1980–89 Model 1 is preferred (p = 2.8e�06) 52 (79%)ECI Entropy Pooled Neither model is significantly preferred (p = 0.1) 194 (54%)ECI Entropy 2000–08 Model 1 is preferred (p = 0.019) 65 (62%)ECI Entropy 1990–99 Model 1 is preferred (p = 0.0039) 66 (65%)ECI Entropy 1980–89 Neither model is significantly preferred (p = 0.18) 39 (59%)ECI HHI Pooled Model 1 is preferred (p = 7.8e�10) 236 (66%)ECI HHI 2000–08 Model 1 is preferred (p = 0.00039) 71 (68%)ECI HHI 1990–99 Model 1 is preferred (p = 3.9e�05) 72 (71%)ECI HHI 1980–89 Model 1 is preferred (p = 1.0e�04) 49 (74%)ECI Log GDP Pooled Model 1 is preferred (p = 4.4e�13) 246 (69%)ECI Log GDP 2000–08 Model 1 is preferred (p = 0.00082) 70 (67%)

Notes: Clarke test compares Model 1 (ECI) with different measures of productive structure (Model 2). The dependent variable is GINI ALL.

Table 4 (continued)

Variable Year Obs. Mean SD Min Max

Voice and accountability All decades 226 0.01 0.98 �2.13 1.67Fitness Index 1963–69 102 0.99 2.59 0.00 15.5Fitness Index 1970–79 108 0.94 2.34 0.00 10.1Fitness Index 1980–89 103 1.00 4.54 0.00 45.6Fitness Index 1990–99 125 1.00 1.40 0.00 7.29Fitness Index 2000–08 128 1.00 1.21 0.00 5.81Fitness Index All decades 566 0.99 2.60 0.00 45.6Shannon Entropy 1963–69 102 2.76 1.26 0.04 5.26Shannon Entropy 1970–79 108 2.91 1.38 0.15 5.25Shannon Entropy 1980–89 103 3.22 1.50 0.33 5.65Shannon Entropy 1990–99 125 3.60 1.39 0.47 5.57Shannon Entropy 2000–08 128 3.50 1.39 0.27 5.51Shannon Entropy All decades 566 3.23 1.42 0.04 5.65Herfindahl–Hirschman 1963–69 102 0.23 0.23 0.01 0.99Herfindahl–Hirschman 1970–79 108 0.23 0.26 0.01 0.96Herfindahl–Hirschman 1980–89 103 0.21 0.24 0.01 0.91Herfindahl–Hirschman 1990–99 125 0.15 0.20 0.01 0.83Herfindahl–Hirschman 2000–08 128 0.17 0.21 0.01 0.91Herfindahl–Hirschman All decades 566 0.19 0.23 0.01 0.99

Notes: Descriptive summary statistics for all macro indicators and all decades’ intervals used in the paper and in the robustness checks presented in theSM. The time period, number of observations, mean, standard deviation, minimum, and maximum are shown.

LINKING ECONOMIC COMPLEXITY, INSTITUTIONS, AND INCOME INEQUALITY 89

Table 7. Cross-section regression results

Dependent variable: GINI ALL

(1) (2) (3) (4) (5) (6)

ECI �4.920*** �3.640*** �6.129*** �3.967*** �5.464***

(1.166) (1.213) (1.172) (1.091) (1.200)ln(GDP PPP pc) 28.753*** 27.002*** 26.477*** 27.364*** 32.493***

(4.978) (5.221) (5.160) (4.994) (4.623)ln(GDP PPPpc)2 �1.648*** �1.625*** �1.580*** �1.572*** �1.776***

(0.311) (0.328) (0.324) (0.313) (0.269)Schooling �1.257*** �1.625*** �0.850** �1.362*** �1.272***

(0.325) (0.330) (0.340) (0.325) (0.331)ln population 1.028** 0.264 0.612 1.303*** 0.832*

(0.477) (0.464) (0.522) (0.492) (0.435)Rule of law �10.587*** �10.759*** �12.962*** �12.343*** �10.030***

(2.509) (2.640) (2.715) (2.576) (2.525)Corruption control 6.602*** 8.087*** 5.046* 8.575*** 5.882**

(2.472) (2.575) (2.664) (2.524) (2.477)Government effectiveness �1.448 (3.100) �3.801 (3.210) �1.657 (3.332) �0.573 (3.227) �0.212 (3.082)Political stability �1.262 �2.055* �1.435 �0.970 �2.392**

(1.122) (1.164) (1.237) (1.168) (1.003)Regulatory quality 5.971*** 5.432** 9.519*** 5.152** 4.941**

(2.231) (2.345) (2.381) (2.319) (2.205)Voice and accountability 3.229** (1.324) 3.644*** (1.390) 3.260** (1.448) 2.344* (1.361) 3.240** (1.339)Constant �89.788*** �61.970*** 34.531*** �89.412*** �65.991*** �106.250***

(22.344) (22.471) (9.433) (23.323) (19.651) (21.661)

Observations 167 167 167 167 167 167R2 0.554 0.503 0.448 0.511 0.540 0.449Adjusted R2 0.522 0.471 0.416 0.479 0.511 0.432Residual std. error 6.632 6.980 7.332 6.922 6.709 7.229

(df = 155) (df = 156) (df = 157) (df = 156) (df = 156) (df = 161)F-statistic 17.490*** 15.759*** 14.135*** 16.284*** 18.346*** 26.273***

(df = 11; 155) (df = 10; 156) (df = 9; 157) (df = 10; 156) (df = 10; 156) (df = 5; 161)

Note: *p < 0.1; **p < 0.05; ***p < 0.01.Pooled OLS regression models for GINI ALL. The regression table explores the effects of economic complexity on income inequality in comparison withother the socioeconomic factors, such as a country’s average level of income, population, human capital, and the institutional variables: corruptioncontrol, government effectiveness, political stability, regulatory quality, and voice and accountability. Column I includes all variables. Columns II–VIexclude blocks of variables to explore the contribution of each group of variables to the full model. The table pools data from two panels, one from 1996 to2001 and another one from 2002 to 2008. If not otherwise indicated, the numbers in parenthesis are standard errors (SEM).

Table 8. Per decade cross-sections with GINI EHII

Dependent variable: GINI EHII

1963–69 1963–69 1970–79 1970–79 1980–89 1980–89 1990–99 1990–99 2000–08 2000–08

ECI �2.465*** �1.819* �2.374** �4.193*** �4.103***

(0.776) (0.918) (1.030) (0.864) (0.797)ln(GDP PPP pc) 5.741 8.231 7.117 9.977* 0.654 �0.036 �0.308 1.371 9.515** 8.842*

(5.213) (5.673) (5.734) (5.695) (4.840) (4.998) (3.310) (3.737) (3.886) (4.511)ln(GDP PPPpc)2 �0.453 �0.701* �0.506 �0.737** �0.072 �0.094 0.022 �0.185 �0.545** �0.603**

(0.345) (0.369) (0.376) (0.366) (0.291) (0.300) (0.200) (0.222) (0.220) (0.256)Schooling �0.743** �0.808** �0.909** �0.911** �0.815*** �1.020*** �0.636*** �1.175*** �0.631** �1.030***

(0.316) (0.347) (0.356) (0.366) (0.304) (0.301) (0.235) (0.235) (0.242) (0.266)ln population 0.191 0.056 0.019 �0.143 0.089 �0.414 0.533 �0.231 0.700** 0.078

(0.398) (0.436) (0.459) (0.463) (0.484) (0.447) (0.348) (0.352) (0.313) (0.335)Constant 26.432 24.250 21.614 16.264 44.467* 60.459** 40.546** 56.499*** �2.417 20.036

(23.463) (25.815) (26.624) (27.181) (24.157) (23.941) (15.465) (17.156) (18.085) (20.384)

Observations 48 48 60 60 66 66 83 83 78 78R2 0.822 0.779 0.676 0.652 0.569 0.531 0.715 0.627 0.683 0.567Adjusted R2 0.800 0.758 0.646 0.627 0.533 0.500 0.696 0.608 0.661 0.543Residual std.error

3.002 3.305 3.965 4.069 4.490 4.646 3.742 4.249 3.687 4.283(df = 42) (df = 43) (df = 54) (df = 55) (df = 60) (df = 61) (df = 77) (df = 78) (df = 72) (df = 73)

F-statistic 38.690*** 37.831*** 22.489*** 25.762*** 15.855*** 17.272*** 38.549*** 32.820*** 31.076*** 23.890***

(df = 5; 42) (df = 4; 43) (df = 5; 54) (df = 4; 55) (df = 5; 60) (df = 4; 61) (df = 5; 77) (df = 4; 78) (df = 5; 72) df = =4; 73)

Notes: *p < 0.1; **p < 0.05; ***p < 0.01.Per decade cross-section regression with GINI EHII. The effect of ECI is negative and significant.

90 WORLD DEVELOPMENT

Table 9. Per decade cross-sections with GINI All

Dependent variable: GINI All

1970–79 1970–79 1980–89 1980–89 1990–99 1990–99 2000–08 2000–08

ECI �2.298 �4.373** �6.294*** �3.989**

�1.824 �1.783 �1.867 �1.689ln(GDP PPP pc) 34.641*** 38.609*** 31.528*** 33.530*** 20.272*** 23.484*** 34.380*** 34.429***

�9.487 �9.013 �8.794 �9.153 �6.589 �6.911 �6.829 �7.013ln(GDP PPPpc)2 �2.122*** �2.435*** �1.866*** �2.127*** �1.037** �1.405*** �1.930*** �2.036***

�0.626 �0.579 �0.55 �0.564 �0.405 �0.414 �0.397 �0.405Schooling �0.856 �0.804 �0.248 �0.406 �0.834* �1.516*** �1.096** �1.427***

�0.562 �0.564 �0.54 �0.561 �0.484 �0.466 �0.474 �0.464ln Population 0.237 0.099 0.52 �0.199 1.017 0.024 0.533 0.006

�0.767 �0.765 �0.768 �0.743 �0.665 �0.632 �0.611 �0.584Constant �95.121** �104.858** �95.314** �82.141* �65.237** �46.523 �108.258*** �90.658***

�44.146 �43.775 �42.134 �43.688 �29.962 �31.208 �31.824 �31.77

Observations 46 46 59 59 89 89 89 89R2 0.542 0.524 0.449 0.387 0.359 0.271 0.403 0.363Adjusted R2 0.485 0.478 0.397 0.341 0.32 0.237 0.367 0.332Residual std. error 6.095 (df = 40) 6.138 7.03 7.35 7.756 (df = 83) 8.220 (df = 84) 7.506 7.708

(df = 41) (df = 53) (df = 54) (df = 83) (df = 84)F-statistic 9.481***

(df = 5; 40)11.293***

(df = 4; 41)8.643***

(df = 5; 53)8.509***

(df = 4; 54)9.301***

(df = 5; 83)7.819***

(df = 4; 84)11.200***

(df = 5; 83)11.955***

(df = 4; 84)

Notes: *p < 0.1; **p < 0.05; ***p < 0.01.Per decade cross-section regression with Gini All. The effect of ECI is negative and significant.

Table 10. Cross-section regression results

Dependent variable: GINI EHII

(1) (2) (3) (4) (5) (6)

ECI �0.040*** �0.036***

(0.007) (0.007)Fitness Index �0.023***

(0.005)Entropy �0.025***

(0.005)HHI 0.146*** 0.058

(0.044) (0.044)ln(GDP PPP pc) 0.067** 0.036 0.086*** 0.065** 0.059* 0.068**

(0.028) (0.029) (0.029) (0.031) (0.032) (0.028)ln(GDP PPPpc)2 �0.004** �0.002 �0.005*** �0.004** �0.004* �0.004**

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)Schooling �0.005*** �0.008*** �0.008*** �0.008*** �0.009*** �0.005***

(0.002) (0.002) (0.002) (0.002) (0.002) (0.002)ln population 0.007** 0.010*** 0.008*** 0.004 0.0001 0.007***

(0.003) (0.003) (0.003) (0.003) (0.003) (0.003)Rule of law �0.013 �0.008 �0.013 �0.017 �0.016 �0.014

(0.013) (0.013) (0.013) (0.014) (0.014) (0.013)Corruption control 0.011 0.009 0.011 0.019 0.027* 0.009

(0.013) (0.014) (0.013) (0.014) (0.014) (0.013)Government effectiveness 0.002 �0.013 �0.007 �0.012 �0.022 0.003

(0.017) (0.017) (0.017) (0.018) (0.018) (0.017)Political stability �0.010 �0.011* �0.014** �0.017** �0.017** �0.011*

(0.006) (0.007) (0.006) (0.007) (0.007) (0.006)Regulatory quality �0.006 �0.006 0.001 �0.0002 �0.012 �0.002

(0.012) (0.013) (0.013) (0.014) (0.014) (0.013)Voice and accountability 0.001 0.009 0.015* 0.011 0.006 0.004

(0.008) (0.008) (0.008) (0.008) (0.009) (0.008)Constant 0.083 0.199 0.132 0.206 0.286** 0.071

(0.130) (0.131) (0.132) (0.138) (0.141) (0.130)

Observations 142 142 142 142 142 142R2 0.717 0.698 0.703 0.667 0.639 0.721Adjusted R2 0.693 0.672 0.678 0.639 0.612 0.695Residual std. error 0.035 0.036 0.035 0.037 0.039 0.034

(df = 130) (df = 130) (df = 130) (df = 130) (df = 131) (df = 129)(continued on next page)

LINKING ECONOMIC COMPLEXITY, INSTITUTIONS, AND INCOME INEQUALITY 91

Table 10 (continued)

Dependent variable: GINI EHII

(1) (2) (3) (4) (5) (6)

F-statistic 29.916*** 27.282*** 28.014*** 23.698*** 23.208*** 27.720***

(df = 11; 130) (df = 11; 130) (df = 11; 130) (df = 11; 130) (df = 10; 131) (df = 12; 129)

Notes: *p < 0.1; **p < 0.05; ***p < 0.01Pooled cross-section regression using different measures of productive structure, using GINI EHII as a dependent variable. ECI, Fitness Index, andEntropy are negatively correlated with Gini, and HHI is positively correlated with Gini. All regression coefficients between Gini and the four measures ofproductive structure are significant.

Table 11. Fixed-effects regression results GINI EHII

Dependent variable: GINI EHII

(1) (2) (3) (4) (5) (6)

ECI �2.189*** �2.302***

(0.555) (0.570)Fitness Index �0.070

(0.075)Entropy �0.067

(0.408)HHI 0.080 �2.054

(2.352) (2.340)ln(GDP PPP pc) �1.541

(2.915)�2.749(3.009)

�3.285(3.044)

�3.171(2.996)

�3.182(2.973)

�1.735(2.925)

ln(GDP PPPpc)2 �0.019(0.181)

0.015(0.188)

0.049(0.189)

0.043(0.187)

0.044(0.186)

�0.002(0.182)

Schooling 1.380*** 1.475*** 1.444*** 1.454*** 1.453*** 1.334***

(0.282) (0.291) (0.295) (0.294) (0.290) (0.287)ln population �2.213**

(1.098)�1.724(1.124)

�1.708(1.142)

�1.671(1.133)

�1.675(1.122)

�2.357**

(1.111)

Observations 335 335 335 335 335 335R2 0.216 0.167 0.164 0.164 0.164 0.219Adjusted R2 0.152 0.117 0.115 0.115 0.116 0.153Residual std. error 12.959***

(df = 5; 235)9.448***

(df = 5; 235)9.242***

(df = 5; 235)9.236***

(df = 5; 235)11.594***

(df = 4; 236)10.917***

(df = 6; 234)

Notes: *p < 0.1; **p < 0.05; ***p < 0.01.Effects of different diversity measures in fixed-effects panel regressions, using GINI EHII.

Table 12. List of the five products with the highest and lowest PGI values during 1995–2008

SITC4 Product Name Product Section PGI

Five products with highest PGI

721 Cocoa Beans Food and live animals 0.506814 Inedible Flours of Meat and Fish Food and live animals 0.5052683 Fine Animal Hair Crude materials, inedible, except fuels 0.5036545 Jute Woven Fabrics Manufactured goods classified chiefly by material 0.4992875 Zinc Ore Crude materials, inedible, except fuels 0.498

Five products with lowest PGI

7259 Paper Making Machine Parts Machinery and transport equipment 0.3347244 Textile Machinery Machinery and transport equipment 0.3367233 Road Rollers Machinery and transport equipment 0.3382120 Raw Furs Crude materials, inedible, except fuels 0.3387252 Paper Making Machines Machinery and transport equipment 0.340

Notes: PGI Ranking: List of the five products with the respective highest and lowest PGI values during 1995–2008.

92 WORLD DEVELOPMENT

Table 14 illustrates that the PGI values for different decadesare highly correlated with each other, and that this correla-tion—as expected—tends to decline over time.

Correlation between PGI and product complexityFigure A2 illustrates a strong and negative correlation

between PGI and the Product Complexity Index (PCI)

Table 13. Descriptive statistics of PGI values for different decades

Time period Mean Std. Min Mmax

1970–79 0.367 0.043 0.285 0.5121980–89 0.383 0.042 0.305 0.5041990–99 0.403 0.039 0.327 0.4962000–08 0.418 0.038 0.337 0.518

Table 14. Correlation coefficient between PGI values from different decades

PGI value 1970–79 1980–89 1990–99 2000–08

1970–79 1.000 0.879 0.734 0.6671980–89 1.000 0.841 0.7461990–99 1.000 0.8692000–08 1.000

Figure A2. Bivariate relationship between the Product Complexity Index (PCI) and the Product Gini Index (PGI) in the (A) 1970–79, (B) 1980–89, (C)

1990–99, and (D) 2000–08.

LINKING ECONOMIC COMPLEXITY, INSTITUTIONS, AND INCOME INEQUALITY 93

(Felipe et al., 2012; Hausmann et al., 2014; Hidalgo &Hausmann, 2009) for different decades. In other words, morecomplex industrial products tend to be associated with lowerlevels of inequality.

APPENDIX. SUPPLEMENTARY DATA

Supplementary data associated with this article can befound, in the online version, at http://dx.doi.org/ 10.1016/j.worlddev.2016.12.020.

ScienceDirectAvailable online at www.sciencedirect.com